In this paper, we present a comparative evaluation of two speech and noise estimation methods commonly used with nonlinear hearing devices: the coherence function used for spectral estimation, and a noise phase-inversion scheme used to perform signal separation. The speech and noise spectra estimated at the output with both methods are compared for normal-hearing subjects and for hearing-impaired subjects using linear and nonlinear hearing aid processing. The spectra are found to be similar except for very low SNR conditions. However, Speech Intelligibility Index (ANSI S3.5-1997 [1]) estimates are relatively unaffected by differences in the spectra obtained with each method.